Fix choose_bitmap_and() so that partial index predicates are considered when

deciding whether a potential additional indexscan is redundant or not.  As now
coded, any use of a partial index that was already used in a previous AND arm
will be rejected as redundant.  This might be overly restrictive, but not
considering the point at all is definitely bad, as per example in bug #2441
from Arjen van der Meijden.  In particular, a clauseless scan of a partial
index was *never* considered redundant by the previous coding, and that's
surely wrong.  Being more flexible would also require some consideration
of how not to double-count the index predicate's selectivity.

diff --git a/src/backend/optimizer/path/indxpath.c b/src/backend/optimizer/path/indxpath.c
index 935494e077854e979ef8d4fdd9f6a1233a5f4d94..ac6de3d82a5c5b86e656329082cdb2ac1242e65d 100644 (file)
--- a/src/backend/optimizer/path/indxpath.c
+++ b/src/backend/optimizer/path/indxpath.c
@@ -9,7 +9,7 @@
  *
  *
  * IDENTIFICATION
- *       $PostgreSQL: pgsql/src/backend/optimizer/path/indxpath.c,v 1.205 2006/05/18 17:12:10 tgl Exp $
+ *       $PostgreSQL: pgsql/src/backend/optimizer/path/indxpath.c,v 1.206 2006/05/18 19:56:46 tgl Exp $
  *
  *-------------------------------------------------------------------------
  */
@@ -616,10 +616,10 @@ choose_bitmap_and(PlannerInfo *root, RelOptInfo *rel, List *paths)
         * non-selective.  In any case, we'd surely be drastically misestimating
         * the selectivity if we count the same condition twice.
         *
-        * XXX is there any risk of throwing away a useful partial index here
-        * because we don't explicitly look at indpred?  At least in simple cases,
-        * the partial index will sort before competing non-partial indexes and so
-        * it makes the right choice, but perhaps we need to work harder.
+        * We include index predicate conditions in the redundancy test.  Because
+        * the test is just for pointer equality and not equal(), the effect is
+        * that use of the same partial index in two different AND elements is
+        * considered redundant.  (XXX is this too strong?)
         *
         * Note: outputting the selected sub-paths in selectivity order is a good
         * thing even if we weren't using that as part of the selection method,
@@ -732,14 +732,14 @@ bitmap_and_cost_est(PlannerInfo *root, RelOptInfo *rel, List *paths)
 /*
  * pull_indexpath_quals
  *
- * Given the Path structure for a plain or bitmap indexscan, extract a
- * list of RestrictInfo nodes for all the indexquals used in the Path.
+ * Given the Path structure for a plain or bitmap indexscan, extract a list
+ * of all the indexquals and index predicate conditions used in the Path.
  *
  * This is sort of a simplified version of make_restrictinfo_from_bitmapqual;
  * here, we are not trying to produce an accurate representation of the AND/OR
  * semantics of the Path, but just find out all the base conditions used.
  *
- * The result list contains pointers to the RestrictInfos used in the Path,
+ * The result list contains pointers to the expressions used in the Path,
  * but all the list cells are freshly built, so it's safe to destructively
  * modify the list (eg, by concat'ing it with other lists).
  */
@@ -777,7 +777,8 @@ pull_indexpath_quals(Path *bitmapqual)
        {
                IndexPath  *ipath = (IndexPath *) bitmapqual;
 
-               result = list_copy(ipath->indexclauses);
+               result = get_actual_clauses(ipath->indexclauses);
+               result = list_concat(result, list_copy(ipath->indexinfo->indpred));
        }
        else
                elog(ERROR, "unrecognized node type: %d", nodeTag(bitmapqual));